GPU Dedicated Servers: How to Choose One for Your Work
A GPU dedicated server is a whole machine with a graphics card only you use. How to match the card to the job, what else to check, and what is allowed on one.
3 min readBy Swati
Reviewed by the AmazingRDP technical team

Seven builds, from entry graphics to a workstation card. See the GPU dedicated servers. See GPU Dedicated Servers from $109.99
A GPU dedicated server is a physical machine in a datacentre with a graphics card, rented to one customer. Nothing is shared: the processor, the memory, the disks and the card are yours, with full administrator access. That is the difference from a GPU RDP, where several users share one card. This page helps you choose a build; the sales details are on the plan page.
Start from the job, not from the card
| Job | What matters in the card |
|---|---|
| Live streaming, video transcoding | A hardware video encoder. Intel's built-in graphics (Quick Sync) and NVIDIA cards with NVENC both qualify; raw 3D power matters little |
| Android emulators | A real graphics card with current drivers; processor cores and memory count as much |
| 3D rendering | Speed and enough video memory for the scene |
| Machine learning | An NVIDIA card (CUDA) and, above all, video memory |
| A desktop that feels like a PC | Any real card |
The classes of card you will meet
Our seven builds, all in Europe, fall into four classes. Models change with stock; the plan page lists what is offered today.
| Class | Good for | Not for |
|---|---|---|
| Entry display cards (such as a GeForce GT 710) | Giving the server a real display adapter | Hardware encoding, rendering, machine learning: such cards lack the encoder and the memory |
| Intel built-in graphics (HD/UHD Graphics) | Video encoding and streaming through Quick Sync, at low cost, beside a strong processor | 3D work, CUDA software |
| Gaming cards (such as a GeForce GTX 1080 with 8 GB) | NVENC encoding, GPU rendering of moderate scenes, emulators, older or small AI models | Large current AI models |
| Workstation cards (RTX class, up to an RTX 6000 Ada) | Heavy rendering, machine learning, anything needing a great deal of video memory | Small budgets |
The point of the table: a build named "GPU" is not automatically right for every GPU job. Compare the card's model with what your software's documentation asks for. Why an encoder may be missing on an entry card is explained in GPU encoding vs CPU encoding, and an older gaming card is assessed in the GTX 1080 as a server card.
The rest of the machine
- Processor: emulators and encoding lean on it heavily. Count cores for the number of things you run at once.
- Memory: machine learning data sets and many emulator instances eat it quickly.
- Storage: NVMe drives; builds with two drives can mirror them (see RAID explained).
- Connection and traffic: streaming around the clock adds up to a lot of data.
- System: Windows as an evaluation edition, or Linux. On Linux you install the card's driver yourself: NVIDIA drivers on Ubuntu.
What is allowed here that is not allowed on shared GPU plans
| GPU RDP (shared) | GPU dedicated server | |
|---|---|---|
| Android emulators | No | Yes |
| Rendering for many hours | No | Yes |
| Streaming all day | No | Yes |
| Your own drivers and software | No | Yes |
| Cryptocurrency mining | No | No. Forbidden on every plan |
The usage policy applies as everywhere else.
What to expect when ordering
- Setup time: a physical server is prepared by hand. In stock, that takes 3 to 4 hours; otherwise up to 24 hours, rarely longer. The workstation-card builds can take several days.
- Upgrades: a different build is a different machine, so changing means a new order and moving your data.
- Refund: 7 days with used days deducted, under the refund policy.
- Questions first: if a particular driver, card feature or program is essential, ask before ordering. How to open a ticket.
The first hour on the server
- Connect and change the password: how to connect.
- Check that the card is present: Task Manager, Performance, on Windows; GPU information on Linux.
- Install or update the graphics driver from the card maker's site.
- Run your real workload for an hour and watch the card's load and memory.
- Secure the server: securing RDP.
Shared or dedicated?
If you need a graphics card for an hour or two of work at a time, the shared plan costs less. For anything that keeps the card busy, a dedicated server is the right tool. The decision in detail: GPU RDP or GPU dedicated server. What graphics cards do beyond games: eight jobs for a GPU.
Questions
What is a GPU dedicated server?
A whole physical server with a graphics card, used by one customer only, with full administrator access.
Which GPU server do I need for streaming?
One whose graphics have a hardware video encoder: Intel Quick Sync or an NVIDIA card with NVENC. Entry display cards may lack it.
Can I run Android emulators on a GPU dedicated server?
Yes. GPU dedicated servers are the plans on which emulators are allowed.
Is mining allowed on a GPU dedicated server?
No. Cryptocurrency mining is forbidden on all plans.